In forestry sector, the remote sensing technology hold a key role on forest inventory and
monitoring their changes. This paper describes the algorithm for detecting deforestation and forest
degradation using high resolution satellite imageries with knowledge-based approach. The main objective
of the study is to develop a practical technique for monitoring deforestation and forest degradation
occurred within the mangrove and swamp forest ecosystem. The SPOT 4, 5, and 6 images acquired in
2007, 2012 and 2014 were transformed into three vegetation indices, i.e., Normalized Difference
Vegetation Index (NDVI), Green-Normalized Difference Vegetation index (GNDVI) and Normalized
Green-Red Vegetation index (NRGI). The study found that deforestation was well detected and identified
using the NDVI and GNDVI, however the forest degradation could be well detected using NRGI, better
than NDVI and GNDVI. The study concludes that the strategy for monitoring deforestation, biomass-based
forest degradation as well as forest growth could be done by combining the use of NDVI, GNDVI and
NRGI respectively.
2. ◼ ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 5, October 2019: 2335-2345
2336
countries due to reduced biodiversity, the extinction of germ plasm sources and the excessive
accumulation of CO2 in the atmosphere, which will eventually lead to significant increases in
the temperature of the earth, and trigger climate change. Some researchers say that the highest
greenhouse gas emissions come from forest conversion into other non-vegetated forest uses.
In developing countries, particularly, forest degradation is also become a fundamental problem.
It has been reported by ITTO [9] in 2000 that the total area of degraded forest in 77 countries
are about 800 million hectares (ha) where 500 million ha of them are degraded from primary to
secondary forest.
In tropical ecosystems, particularly in the region with high economic growth, the forest
ecosystems are suffering from many other land use pressures and conversion. To anticipate
the unauthorized forest conversion, then the automated, practical and fast monitoring technique
is required to provide accurate and comprehensive estimation by ecosystem type and its
geographic location. In this study, the authors focused on the development of deforestation and
forest degradation monitoring techniques. The monitoring system for detecting forest condition
could be found in several studies such as Maselli [10-13]. At global level, it was mentioned that
NOAA-AVHRR-based NDVI closely correlated with forest ecosystem variation [10]. Other
studies [14, 15] used NDVI as an indicator of forest degradation. At the tree-crown level, the use
of spectral indices had been examined for detecting defoliation of Eucalypts [16] and the use of
NDVI trend to detect forest cover changes [17]. In some studies, it were pointed out that some
methodological and sensors were implementable for monitoring of the forest degradation using
remote sensing in tropical forest [18-23]. The use of satellite data of LISS III for monitoring
deforestation and degradation had also been reported [24]. In more detailed level, Da [25]
reported the use of remote sensing data to asses the forest dynamic change at household level
data. The use of NDVI for monitoring land degradation in arid environment has been also
examined [26]. Now, the use of high spatial resolution and hyper-spectral imageries for
detecting change is also a highlighted issue [27, 28].
In change detection studies that have been done previously, none of them has a study
related to the development of algorithms to detect forest cover changes, especially deforestation
and forest degradation. While remote sensing technology is developing very rapidly, particularly
the ability of spatial resolution and temporal resolution. Spatial resolution has arrived at
the ability to detect objects up to sub-meters, while the temporal resolution of high spatial
resolution sensor had been capable to acquire the data at every 1 to 3 days (revisit time). In this
regard, the authors examined the algorithm for detecting deforestation and forest degradation.
In this study, deforestation detection was analyzed using conventional vegetation indices such
as NDVI and GNDVI, while detection of forest degradation was done with the modified
vegetation index. The main objective of the study is to provide a practical technique for
monitoring deforestation and forest degradation, particularly within the mangrove and swamp
forest ecosystem. This study is expected to be a factor in the development of forest cover
change detection methods, such as deforestation and forest degradation, especially on
mangrove forest ecosystems which became one of the important ecosystems in the tropics.
This algorithm will help detect changes in the mangrove region semi-automatically using
high-resolution satellite imagery.
2. Research Method
2.1. Date and Study Site
For ground data collection, the study was conducted from February 2015 to mid 2016,
at two different sites, in Kubu Raya Regency, West Kalimantan as shown in Figure 1.
Geographically, the study sites is located between 109o18’0” E and 109o42’0” E and between
0o30’00” and 0o55’00”. The data processing and analysis was carried out at the Remote
Sensing and GIS Laboratory, Forest Management Department, Faculty of Forestry,
Bogor Agricultural University.
2.2. Data, Software and Hardware
The main data used in this study are medium-resolution image, i.e., SPOT 4 and 5
acquired in 2007 and 2012 as well as high-resolution image SPOT 6 recorded in 2014, covering
mangrove and swamp forest ecosystem in Kubu Raya District, West Kalimantan Province.
The SPOT 6 data has 6 m spatial resolution, whereas SPOT 4 and 5 images have 10 m spatial
3. TELKOMNIKA ISSN: 1693-6930 ◼
Algorithm for detecting deforestation and forest degradation using... (I Nengah Surati Jaya)
2337
resolution. The other supporting data that are used were field observation data including forest
condition, standing stock, geographic coordinate point, year of land clearing, terrestrial-based
standing stock, administrative boundary and the base map of West Kalimantan. The algorithm
for detecting the deforestation and forest degradation, starting from the image pre-image
processing to image processing was used ERDAS imagine software and operated at
the desktop computer. The establishment image vegetation indices, examination of
deforestation criteria, degradation and growth were done by utilizing the capability of ERDAS
"modeler", while the statistical analysis for each synthetic image was processed using
the statistical functions of the Microsoft excel.
2.3. Criteria for Detecting Deforestation between 2007, 2012 and 2014
Observations of deforestation on mangrove forests has been observed using both
the visual interpretation and digital analysis. The vegetation index derived from NIR and R
bands could enhance the contrast between cover classes experiencing the biomass and/or
chlorophyll changes. The use of spatial technology for monitoring biomass or carbon stock
could be found in the work of Nyamugama and [29]. In this study, the deforestation information
was derived by examining several threshold values for “bare-land” and “forest cover” that
express “the changes” and “no change”. The algorithm for detecting deforestation was done by
using “knowledge-based" approach with multilevel slice classification method. The forest and
land cover classes, such as mangrove forests, swamp forests, shrubs/bushes, Nypahs, water
bodies, barren land, clouds and cloud shadows were derived in the image of 2007, 2012
and 2014.
To distinguish between vegetation cover and non-vegetation cover, the study examined
the used of the standard vegetation index and it modifications. There are three indices were
examined, namely the Normalized Differenced Vegetation Index (NDVI), Green Normal
Differentiated Vegetation Index (GNDVI) and Normalized Green-Red Vegetation Index (NRGI),
computed with the following formula:
NDVI = (NIR-R)/(NIR+R) (1)
GNDVI = (NIR-G)/(NIR+G) (2)
NRGI = (G-R)/(G+R) (3)
where NIR is infrared R is red band and G is green band. Therefore, the first step of this
deforestation detection is to separate water bodies, bare soil and vegetation by examining
the most optimal threshold value of NDVI (THN) at each pixel values of NDVI image (X1):
If X1 < THN1 Then “Water body”
Else if X1 > THN1 && X1 < THN2 then “Bare land”
Year 2012: NRGIMangrove> NRGISwamp Forest> NRGINypa> NRGIBushes> NRGIwater bodies>
NRGIbare land
furthermore, this study examine the threshold value of GNDVI (THG) at every pixel value of
GNDVI (X2) with the following decision rules:
If X2 < THG1 && X1 = “water body” Then “water body”
Else if X2 > THG2 && X1 = “vegetation “Then “vegetation”
Else “bare land/cloud/cloud shadow”
If X2 > THG1 && X1 > THN1 Then “calculate biomass change or volume change”
then, deforestation and afforestation detections from two difference dates, i.e. DT1 for date 1
and DT2 for date 2 are formulated using the following formula:
If DT1 = “Forest” && DT2 = “Bare land” Then “Deforestation”
Else if DT1 = “Bare land && DT2 = “Forest” Then “Afforestation”
if DT1 = DT2 Then “No-change”
Else “undefined”
4. ◼ ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 5, October 2019: 2335-2345
2338
the “undefined” changes are defined as other changes including cloud and cloud shadow
classes. The change from a water body to bare land or vice versa is frequently due to
the influence of the tides or the season (dry or rainy) at the acquisition time.
2.4. The Forest Degradation Criteria and Regrowth
Forest degradation is defined as changes in forest quality in terms of decreasing stand
volume, forest biomass and/or biological and non-biological environment quality. Many studies
have suggested that remote sensing is able to provide accurate information regarding
the biomass and volume of forest stands. The forest degradation examined in this study was
defined as a forest having a degraded or decreased biomass volume. If the change is negative
beyond the threshold limit then it is called “degradation”, otherwise the positive change
is called "growth". The small negative changes that belongs within the threshold limit is called as
“somehow degradation”, conversely, the small positive change is referred to
as “somehow growth”. Mathematically detected forest degradation and standing growth
are as follows:
If (BT2 -BT1) > -1 && (BT2 -BT1) < 0 && Then “somehow degradation”
Else If (BT2 -BT1) < -1 Then “degradation”
Else If (BT2 -BT1) > 0 && (BT2 -BT1) < 1 Then “somehow growth”
Else If (BT2 -BT1) > TH Then “growth”
Else “No-change”
An area is categorized as degraded, or otherwise as grows if the biomass changes are
more than or equal to 15% of the original biomass volume. The BT2 and BT1 values are
the biomass volume values in the second year (T2) and the previous year (T1) obtained using
the biomass estimation model using the NRGI variable. The estimation model was already
developed in our previous researches.
3. Results and Analysis
3.1. The Separability Analysis of Forest and Land Covers
In monitoring deforestation and forest degradation, consideration of the spectral
capability of each band becomes a very crucial task. Referring to the separability analysis using
the original band of SPOT 4 and 6 images, namely Green, Red and NIR, it is recognized that
the high inter-class separability are occurred between vegetation with bare land, vegetation with
water bodies and or between water bodies with bare land. This is in line with the study results
of [30], where the NIR and Red band hold a significant role on discriminating green vegetation
and burnt vegetation. The separability analysis on the SPOT and TM images could also be
found in [31].
In the 2007 image, the low separability (less than 1600) are only between shrubs and
swamp forests, whereas for 2012 imagery, low separation was occurred between mangrove and
shrubs. In the 2014 image, low class-to-class separation occurs between cloud and water
bodies; as well as between mangrove forests and swamp forest and between bare land and
shrubs. Based on the inter-class separation approach, it was found that the inter-class
separation among 8 classes, using the original data (the R, G and NIR bands), the average of
Transformed Divergence (TD) are 1951, 1909 and 1903, for 2007, 2012 and 2014 respectively.
However, average of separation increases when the synthetic band NDVI, GNDVI and NRGI
are applied having TD around 1986, 1968 and 2000 as shown in Table 1. From the individual
separation of using the original bands, the separation classes between water bodies and cloud
shadows is very low at the 2014 images. In contrast, using the vegetation index, the inter-class
separation was improved significantly, and all classes had an inter-class separation
above 1600.
3.2. The Spectral Analysis of NDVI, GNDVI and NRGI
In general, the use of synthetic images using NDVI, GNDVI and NRGI images could
increase the inter-class separation, particularly between vegetation and non-vegetation classes
(such as water bodies, bare soils, clouds and cloud shadows). Theoretically, the dense
vegetation will have high value of NDVI, GNDVI and NRGI, close to one, whereas the water
body value is close to minus one. From land spectral analysis at the 2007, 2012 and 2014
5. TELKOMNIKA ISSN: 1693-6930 ◼
Algorithm for detecting deforestation and forest degradation using... (I Nengah Surati Jaya)
2339
acquisition date as shown in Figures 1-3, that the NDVI and GNDVI improve separation
between water bodies and vegetation. At all dates, consistently, NDVI shows its superiority to
GNDVI in distinguishing vegetation cover classes, water bodies, clouds, cloud shadows and
bare soils. However, when viewed from the separability between the vegetation cover class and
the water body, it is seen that GNDVI gives wider average values compared to NDVI. In other
words, to distinguish water bodies, bare land and vegetation as well as cloud or cloud shadow,
then NDVI is better than GNDVI. The ratio between NIR and R in NDVI increases
the separability between classes of water bodies, bare land and vegetation. While GNDVI
shown its superiority particularly in classifying vegetation and non-vegetation classes.
The GNDVI is a modification of NDVI where the Red (R) band is replaced with Green (G).
Table 1. Average of TD and Number of class-pair with TD less than 1600 using
original band and synthetic band of SPOT 4, 5 and 6
Year
TD_avg with
original band
Number of
class-pair with
TD < 1600
TD_avg with NDVI
GNDVI, NRGI
Number of
class-pair with TD
< 1600
2007 1951 1 1986 0
2012 1909 3 1969 0
2014 1903 1 2000 0
The NDVI, GNDVI and NRGI bands have unique brightness values patterns. The NDVI
and GNDVI images are superior on distinguishing the water bodies and non-water bodies than
NRGI. To distinguish shadows from cloud shadows, the NDVI provides better separation than
GNDVI. The NRGI could not classify vegetation and non-vegetation cover accurately.
Furthermore, the GNDVI image easily separate water bodies and non-water body classes,
better than NDVI. This study shows that the NRGI images are not sensitive to the vegetation
and non-vegetation classes such as bare land, water bodies, clouds and cloud shadows.
However, as shown in Figures 1-3, the NRGI of 2007, 2012 and 2014 consistently gradation of
biomass volume classes. The observations in Figures 1-3 show that NRGI could distinguish
variations of forest biomass classes or forest potential classes, better and more consistent than
either GNDVI or NDVI. Based on the NRGI sequence, the order of NRGI and biomass values is
as follows:
NRGImgrv > NRGIswf > NRGIsbh > NRGINypa
this is in line with the sequence of biomass volume:
Mangroves > swamp forest > bush / shrub > Nypa.
based on the above considerations, then the algorithm for deforestation and degradation
detection using synthetic channels could be formulated as follows:
a) The threshold value of NDVI could be used to classify vegetation and non-vegetation
covers, such as vegetation, water bodies, bare land, cloud and cloud shadow categories.
Based on the value of its separation, the classes could be easily distinguished. NDVI easily
differentiate cloud shadows and water bodies, where these two classes are difficult to
separate on the original channel as shown in Figures 1-3. The sequence of NDVI values
from dates 2007, 2012 and 2014 in this study are:
NDVI Vegetation> NDVI cloud shadow> NDVI clouds / bare land > NDVIwater bodies
b) In the GNDVI images, the separation between the vegetation and non-vegetation classes is
similar with those obtained from the NDVI. However, if it is examined precisely,
the separation between water bodies and vegetation using GNDVI is better than NDVI.
Thus, water body detection could be well assessed using GNDVI values, better than NDVI.
This concludes that the GNDVI threshold is useful to differentiate the class of water bodies
and non-water body classes.
6. ◼ ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 5, October 2019: 2335-2345
2340
c) Particularly for vegetation sub-classes, the NRGI provides consistent and accurate
estimation of biomass and volume contents. From the data time series examined, the NRGI
values on the 2007, 2012 and 2014 images consistently show that the sequence of NRGI
values as follows:
NRGIMangrove> NRGISwamp Forest> NRGIBushes/Nypa> NRGIwater body
d) Following the classification using NDVI the GNDVI, the NRGI image could be used to
detect forest type and other cover classes such as mangrove, swamp forests
and shrubs/nypa. In this study nypa and bush could not be well distinguished, because this
non-woody vegetation, have a similar biomass content. From the image of 2007, 2012
and 2014, the study found the ability of NRGI in determining stand biomass having
the sequence as follows:
Year 2007: NRGIMangrove> NRGISwamp Forest> NRGIBushes> NRGINypa> NRGIwater bodies>
NRGIbare land
Year 2012: NRGIMangrove> NRGISwamp Forest> NRGINypa> NRGIBushes> NRGIwater bodies>
NRGIbare land
Year 2014: NRGIMangrove> NRGISwamp Forest> NRGIBushes> NRGINypa> NRGIwater bodies>
NRGIbare land
all the NGRI sequence as mentioned above then could be summarized as follows:
Potential mangroves> swamp forest potential> potential shrub / bushes> Nypa potential.
(a) (b)
(c)
Figure 1. The values of NDVI, GNDVI and NRGI for each cover class in (a) 2007
(b) 2012 and (c) 2014
The wetness of soil background condition at the sparse vegetation either in mangrove or
swamp forest ecosystem would influence the value of the vegetation index. As shown in this
study, the NRGI value of shrub is higher than Nypa for the year 2007 and 2014, whereas for
the year 2012 Nypa is larger than shrub. The successful use of vegetation indices for
recognizing land cover classes in wetlands could be found in [12, 13]. It was also found an
index associated with drought are Modified Vegetation Water Supply Index (MVWSI) and
relative precipitation index (RPI) [11]. It is recognized that MVWSI has simpler calculation and
provide more stable result, hence more easily realized in application.
7. TELKOMNIKA ISSN: 1693-6930 ◼
Algorithm for detecting deforestation and forest degradation using... (I Nengah Surati Jaya)
2341
(a) (b) (c) (d)
Figure 2. The appearance of (a) NDVI, (b) GNDVI, (c) NRGI and
(d) Composite MIR-NIR-R image of SPOT 6
As shown in Figure 2, each index provides various levels of contrast. The NDVI and
GNDVI show similar object variations, both capable to distinctly distinguish between water
bodies and vegetation as shown in Figures 2 (a) and (b). However, GNDVI gives better
separation between water bodies and vegetation compared to the NDVI. Yet, the NDVI
produces better separation than GNDVI in terms of increasing the separability between the bare
land and clouds as well as cloud shadows. In Figure 2 (c) water bodies have various variable
NRGI values ranging from very dark to gray (compare with Figures 4 (a) and 4 (b)). In other
words, the NRGI is not able to clearly identify the water bodies and non-water body. As shown,
there is a stream with dark tones and bright tones, and this might contribute to misclassification.
However, the NRGI could consistently differentiate the sequence of biomass potentials for each
forest cover class such as mangroves, swamp forests, vegetation in ecotone (shrub and Nypa).
As the findings discussed above, we may conclude that the vegetation index values
obtained using NDVI, GNDVI and NRGI have a potential capability to detect deforestation and
forest degradation. The selection of threshold limits of each threshold could be done by two
methods, namely by the average method and nearest neighbor method. The order of NDVI
values by land cover from 2007 to 2014 is presented in Tables 2-4.
Furthermore, based on the threshold values of each vegetation index examined, a
decision-making flow diagram for detecting deforestation and forest degradation is developed as
shown in Figure 3. Figure 4 shows one example of the forest degradation detection based on
biomass stock. The biomass estimation projection model used was the NRGI-based model
already developed our previous research. Forest degradation due to the decrease in biomass
content or growth due to increased biomass could be detected well using the algorithm
examined in this study, particularly when the image is in excellent quality (free of clouds, cloud
shadows, no stripping/banding or haze). The presence of haze on one image may lead to
misclassification and mis-identification. The presence of haze in the second-year causes
changes to be detected as degradation, otherwise the presence of haze or thin clouds in
previous imagery will cause the change as growth. Detection of changes that resemble "noise"
in the river or shoreline less than 2 pixels is mostly due to miss-registration.
Table 2. Threshold Values using Average and Nearest Method for NDVI in
2007, 2012 and 2014) for Each Class
Land cover NDVI 07 Land cover NDVI 12 Land cover NDVI 14
Mangrove 0.7655 Mangrove 0.7619 Mangrove 0.7737
Swamp forest 0.7008 Nypa 0.7378 Nypa 0.7472
Nypa 0.6852 Swamp forest 0.7135 shrub 0.7152
Shrub 0.6728 Shrub 0.7101 Swamp forest 0.6952
Thr-avg 0.6267 0.6101 0.6418
Thr-nnb 0.6100 0.5998 0.6230
Cloud shadow 0.5473 Cloud shadow 0.4895 Cloud shadow 0.5508
Thr-avg 0.4290 0.3139 0.4851
Thr-nnb 0.4763 0.3562 0.5062
Bare land 0.4054 Bare land 0.2228 Cloud 0.4615
Cloud 0.2159 Cloud 0.0539 Bare land 0.3771
Thr-avg (0.0076) (0.0003) 0.1555
Thr-nnb (0.0550) (0.0425) 0.1344
Water body (0.3259) Water body (0.1389) Water body (0.1083)
Remarks: Thr-avg: average-based threshold; Thr-nnb: nearest-neighbour threshold
8. ◼ ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 5, October 2019: 2335-2345
2342
Table 3. Threshold Values using Average and Nearest Method for GNDVI in
2007, 2012 and 2014) for Each Class
Land cover GNDVI07 Land cover GNDVI12 Land cover GNDVI14
Mangrove 0.5897 Mangrove 0.5874 Mangrove 0.6078
Swamp forest 0.5349 Nypa 0.5743 Nypa 0.5905
Nypa 0.5213 Shrub 0.5423 Shrub 0.5389
Shrub 0.5034 Swamp forest 0.5288 Swamp forest 0.5063
Thr-avg 0.4264 0.3819 0.4562
Thr-nnb 0.4094 0.3672 0.4289
Cloud shadow 0.3154 Cloud shadow 0.2056 Cloud 0.3515
Bare land 0.2793 Bare land 0.1715 Cloud shadow 0.2981
Cloud 0.2446 Cloud -0.0259 Bare land 0.2115
Thr-avg -0.1192 -0.1112 -0.0200
Thr-nnb -0.1368 -0.1827 -0.0578
Water body -0.5183 Water body -0.3395 Water body -0.3271
Table 4. Threshold Values using Average and Nearest Method for NRGI in
2007, 2012 and 2014) for Each Class
Land cover NRGI07 Land cover NRGI12 Land cover NRGI14
Mangrove 0.3206 Mangrove 0.3160 Mangrove 0.3132
Cloud shadow 0.2802 Cloud shadow 0.3157 Cloud shadow 0.3023
Swamp forest 0.2654 Swamp forest 0.2966 Swamp forest 0.2916
Thr-avg 0.2722 0.2938 0.2930
Thr-nnb 0.2608 0.2901 0.2892
Shrub 0.2562 Nypa 0.2837 Shrub 0.2869
Nypa 0.2549 Shrub 0.2728 Nypa 0.2805
Water body 0.2315 Water body 0.2105 Water body 0.2268
Bare land 0.1422 Cloud 0.0797 Bare land 0.1800
Cloud -0.0303 Bare land 0.0533 Cloud 0.1314
Figure 3. The flow diagram for detecting deforestation and biomass-based forest degradation
using remote sensing approach
Evaluate
GNDVI
9. TELKOMNIKA ISSN: 1693-6930 ◼
Algorithm for detecting deforestation and forest degradation using... (I Nengah Surati Jaya)
2343
(a)
(b)
Figure 4. Example of detecting forest degradation between two dates
(a) 2007-2012 and (b) 2012-2014
4. Conclusion
The study concludes that the algorithm for detecting deforestation and forest
degradation should include the use of NDVI, GNDVI and NRGI derived from SPOT 4/5 and/or
SPOT 6. The algorithm is started by detecting deforestation using both NDVI and GNDVI,
respectively, then followed by using the NRGI. Deforestation could be well detected and
identified when the NDVI and GNDVI are used respectively, but it could not be accurately
identified when NRGI is applied. Conversely, forest degradation is only well identified using
the NRGI index. The study found that NRGI image has advantages over NDVI and GNDVI in
terms of estimating the above ground biomass and is recommended for use in monitoring
degradation and standing growth. The NDVI and GNDVI indices are not sensitive to changes in
biomass or standing forest volume, so those two indices could not be used to detect
degradation and/or growth of stands accurately.
Acknowledgement
This article had developed under financial support of the collaborative project among
the Academy of Forest Inventory and Planning, State Forestry Administration, the People
Republic of China; Indonesian Society for Remote Sensing and Faculty of Forestry,
Bogor Agricultural University on the Development of Joint Forest Resources Monitoring
Methodology in Southeast Asia in 2016. The authors express their high appreciation to
10. ◼ ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 5, October 2019: 2335-2345
2344
everyone who has supported the establishment of this paper, including the anonymous reviewer
who criticized the initial draft for improvement.
References
[1] Souza CM, Siqueira JV, Sales MH, Fonseca AV, Ribeiro JG, Numata I, Cochrane MA, Barber CP,
Roberts DA, Barlow J. Ten-Year Landsat Classification of Deforestation and Forest Degradation in
the Brazilian Amazon. Remote Sens. 2013; 5: 5493-5513.
[2] Alves DB, Pérez-Cabello F, Mimbrero MR. Land-Use and Land-Cover Dynamics Monitored by NDVI
Multitemporal Analysis in A Selected Southern Amazonian Area (Brazil) For the Last Three Decades.
The International Archives of The Photogrammetry, Remote Sensing and Spatial Information
Sciences, Volume Xl-7/W3, 2015 36th
International Symposium on Remote Sensing of Environment.
Berlin. 2015.
[3] Hirschmugl M, Steinegger M, Gallaun H, Schardt M. Mapping Forest Degradation due to Selective
Logging by Means of Time Series Analysis: Case Studies in Central Africa. Remote Sens. 2014; 6:
756-775.
[4] Hudak AT, Wessman CA. Deforestation in Mwanza District, Malawi, from 1981 to 1992, as
determined from Landsat MSS imagery. Applied Geography. 2000; 20:155–175.
[5] Wu YH. Investigation of deforestation in East Africa on regional scales. Master’s thesis.Stockholm:
Department of Physical Geography and Quaternary Geology. Stockholm University. 2011.
[6] Van Marle MJE, van der Werf GR, de Jeu RAM, and Liu YY. Annual South American forest loss
estimates based on passive microwave remote sensing (1990–2010). Biogeosciences. 2016; 13:
609–624.
[7] Herold M, Román-Cuesta RM, Heymell V, Hirata Y, Van Laake P, Asner GP, Souza C, Avitabile V,
MacDicken K. A review of methods to measure and monitor historical carbon emissions from forest
degradation Unasylva. 2011; 62(2): 238.
[8] Bright BC, Hicke JA, Hudak AT. Estimating aboveground carbon stocks of a forest affected by
mountain pine beetle in Idaho using lidar and multispectral imagery. Remote Sensing of
Environment. 2012; 124: 270–281.
[9] ITTO. ITTO Guidelines for restoration, management and rehabilitation of degraded and secondary
tropical forest. ITTO Policy Development Series No. 13. Yokohama, Japan, International Tropical
Timber Organization Yokohama, Japan, International Tropical Timber Organization (also available at
www.itto.int/ policypapers_guidelines/). 2002.
[10] Maselli F. Monitoring forest conditions in a protected Mediterranean coastal area by the analysis of
multiyear NDVI data IBIMET-CNR, Institute of Bio-Meteorology, Piazzale delle Cascine 18, 50144,
Florence, Italy. Remote Sensing of Environment. 2004; 89: 423–433.
[11] Men-xin W, Hou-quan LU. A modified vegetation water supply index (MVWSI) and its application in
drought monitoring over Sichuan and Chongqing, China. Journal of Integrative Agriculture.
2016; 15(9): 2132–2141.
[12] Zhao L, Dai A, Dong B. Changes in global vegetation activity and its driving factors during
1982–2013. Agricultural and Forest Meteorology. 2018; 249: 198–209.
[13] Dutta D, Kundu A, Patel NR, Saha SK, Siddiqui AR. Assessment of agricultural drought in Rajasthan
(India) using remote sensing derived Vegetation Condition Index (VCI) and Standardized
Precipitation Index (SPI). The Egyptian Journal of Remote Sensing and Space Sciences. 2015;
18: 53–63.
[14] Meneses-Tovar CL. 2011. NDVI as indicator of degradation. Unasylva 238. 2011; 62(2): 39–46
[15] FAO. Analysis of the normalized differential vegetation index (NDVI) for the detection of degradation
of forest coverage in Mexico 2008–2009: case studies on measuring and assessing forest
degradation. Forest Resources Assessment Working Paper. 2009: 173
[16] Barry KM, Stone C, Mohammed CL. Crown-scale evaluation of spectral indices for defoliated and
discoloured Eucalypts. International Journal of Remote Sensing. 2008; 29(1): 47–69.
[17] Krakauer NY, Lakhankar T, Anadón JD. Mapping and Attributing Normalized Difference Vegetation
Index Trends for Nepal. Remote Sens. 2017; 9: 986.
[18] Lambin EF. Monitoring forest degradation in tropical regions by remote sensing: some
methodological issues. Department of Geography, University of Louvain, place Louis Pasteur.
Global Ecology and Biogeography. 1999; 8: 191–198.
[19] Violini S. Deforestation: Change Detection in Forest Cover using Remote Sensing. Master thesis.
Institute, CONAE. Argentina. Master’s in emergency early Warning and Response Space
Applications. Mario Gulich. 2013.
[20] Langer AJ. Monitoring Tropical Forest Degradation and Deforestation in Borneo, South East Asia.
PhD dissertation. Germany: GeoBio Center of the Ludwig-Maximilian-University Munich. 2009.
[21] Buhalău T. Detecting clear-cut deforestation using Landsat data: A time series analysis of remote
sensing data in Covasna County, Romania between 2005 and 2015. Master thesis. Sweden:
Department of Physical Geography and Ecosystem Science, Lund University. 2016.
11. TELKOMNIKA ISSN: 1693-6930 ◼
Algorithm for detecting deforestation and forest degradation using... (I Nengah Surati Jaya)
2345
[22] Bajracharya S. Community Carbon Forestry: Remote Sensing of Forest Carbon and Forest
Degradation in Nepal. Master thesis. Enschede: The Netherlands. International Institute for
Geo-Information Science and Earth Observation. 2008.
[23] Neba SG. Assessment and prediction of above-ground biomass in selectively logged forest
concessions using field measurements and remote sensing data: case study in South East
Cameroon. Master thesis. Helsinki: Forest Ecology and Management University of Helsinki,
Department of Forest Sciences, Viikki Tropical Resources Institute VITRI. 2013.
[24] Kumar P, Rani M, Pandey PC, Majumdar A, Nathawat MS. Monitoring of Deforestation and Forest
Degradation Using Remote Sensing and GIS: A Case Study of Ranchi in Jharkhand (India). Birla
Institute of Technology. Report number. 2010: 2(4).
[25] Da Ponte E, Mack B, Wohlfart C, Rodas O, Fleckenstein M, Oppelt N, Dech S, Kuenzer, C.
Assessing Forest Cover Dynamics and Forest Perception in the Atlantic Forest of Paraguay,
Combining Remote Sensing and Household Level Data. Forest. 2017; 8(10):389.
[26] Eckert S, Hüsler F, Liniger H, Hodel E. Trend Analysis of MODIS NDVI Time Series for Detecting
Land Degradation and Regeneration in Mongolia. Journal of Arid Environments. 2015; 113: 16−28.
[27] Zhang P, Zhiyong LV, Gao L. A New Framework of the Unsupervised Classification for
High-Resolution Remote Sensing Image. TELKOMNIKA Indonesian Journal of Electrical
Engineering. 2012; 10(7): 1746−1755.
[28] Luo R, Pi Y. Discriminative Supervised Neighborhood Preserving Embedding Feature Extraction for
Hyperspectral-image Classification. TELKOMNIKA Indonesian Journal of Electrical Engineering.
2012; 10 (5): 1051−1056.
[29] Nyamugama A, Kakembo V.2014. Estimation and Monitoring of Aboveground carbon stocks using
Spatial Technology. S Afr J Sci. 2015; 111(9/10): 1-7.
[30] Huang H, Roy DP, Boschetti L, Zhang HK, Yan L, Kumar SS, Gomez-Dans J, Li J. Separability
Analysis of Sentinel-2A Multi-Spectral Instrument (MSI) Data for Burned Area Discrimination. Remote
Sens. 2016; 8: 873.
[31] Lu D, Batistella M, Moran EF, Miranda EE. A Comparative study of Landsat TM and SPOT HRG
Images for Vegetation Classification in the Brazilian Amazon. Photogrammetric Engineering &
Remote Sensing. 2008; 74 (3): 311–321.